Recommended Ogg Vorbis: Difference between revisions

From Hydrogenaudio Knowledgebase
m (→‎Reference Builds: spelling correction)
Line 93: Line 93:
These are highly optimized encoders developed by the '''[http://homepage3.nifty.com/blacksword/index_e.htm Ogg Vorbis Acceleration Project]''' codenamed [[Lancer]]. They are much faster than the standard binary builds having negligible to nearly no effects on audio quality. These include sped-up routines, i.e ''mdct.c''
These are highly optimized encoders developed by the '''[http://homepage3.nifty.com/blacksword/index_e.htm Ogg Vorbis Acceleration Project]''' codenamed [[Lancer]]. They are much faster than the standard binary builds having negligible to nearly no effects on audio quality. These include sped-up routines, i.e ''mdct.c''


''''These optimized encoders where are one point rapidly changing, as BlackSword found new ways to accelerate, and in the process uncovers new bugs. Please check the “Lancer homepage” link below for the last suite release. They have not been updated since 2006. Developers are better off building their own compiles by downloading the source code from the website above and optmizing them for different architectures. The builds on the website above use old AoTuV builds proceed with caution if you decide to use these.  ''''
'''These optimized encoders where are one point rapidly changing, as BlackSword found new ways to accelerate, and in the process uncovers new bugs. Please check the “Lancer homepage” link below for the last suite release. They have not been updated since 2006. Developers are better off building their own compiles by downloading the source code from the website above and optmizing them for different architectures. The builds on the website above use old AoTuV builds proceed with caution if you decide to use these.  '''


On '''[http://homepage3.nifty.com/blacksword/index.htm Lancer homepage]''' you can find older versions of Oggenc, [[OggDropXPd]], and Dynamic Link Libraries, with optimizations for SSE, SSE2, SSE3 and multi-threading instruction sets.
On '''[http://homepage3.nifty.com/blacksword/index.htm Lancer homepage]''' you can find older versions of Oggenc, [[OggDropXPd]], and Dynamic Link Libraries, with optimizations for SSE, SSE2, SSE3 and multi-threading instruction sets.

Revision as of 05:09, 26 June 2010

Vorbis, being a continuously developed standard, improves all the time. Plus, being an open (i.e. patent-free) standard, it has many 'third-parties' that contribute, discuss, and work to improve the project.

Here you can find some guidelines on which utilities to use, and what settings will provide you with the best quality.

History

(For a highly detailed description of Vorbis history, check out OggZealot's Ogg Vorbis Historic where Monty also adds a few more details too).

(Ogg) Vorbis reached version 1.0 in July 2002. It is the official Xiph.org encoder (the one you get from vorbis.com). HA codec developer, Garf, did his own tunings, based on version 1.0 to produce GT3b1 and GT3b2. Both encoders showed improved pre echo handling for q values of 5 to 10. It was later judged in an internal listening test that GT3b1 was the better of the two. There was a minor bugfix update earlier that year in March, which only appeared in the CVS at Xiph.org. This consisted of very minor bug fixes, which do more to correct odd problems that may occur rather than improving quality, including (garbled noise output and gaps in streams). This was referred to as Post 1.0 CVS. Quality problems that mainly affected low bitrates were later addressed in a new bugfix (1.0.1) that was released.

Post 1.0.1 CVS was released late December 2003 by Monty at Xiph, and includes a true CBR template. In order to simplify the situation where we had two encoders (1.0.1 and GT3b1), OggDropXPd developer John33 merged the sources to give us GT3b2. Once the 128 kbps multiformat test was completed, Aoyumi's aoTuV Vorbis tuning was determined to be the best Vorbis encoder. After the success of aoTuV beta 2 encoder, Xiph.Org merged their tunings into the official CVS branch to produce the long-awaited Vorbis 1.1. Aoyumi's later release of aoTuV beta 4 encoder as of November 2005 significantly improves Vorbis' quality while increasing the compression ratio slightly.

Later on, Aoyumi released aoTuV beta 4.5 (later bugfixed with aoTuV beta 4.51) in December 2005, which improves low bit-rate quality even more. After extensive testing by Hydrogenaudio enthusiasts, aoTuV beta 4.51 is re-branded to aoTuV Release 1, and it became the then recommended encoder of Hydrogenaudio. Aoyumi keeps tuning aoTuV. The newest aoTuV encoders are the aoTuV Beta 5 releases. Peer-review by Hydrogenaudio enthusiasts finally decided in June 2007 that aoTuV Beta 5 are now the recommended Vorbis encoders of Hydrogenaudio.

If you are not interested in the latest “bleeding edge” AoTuV improvements, you are welcome to use the latest release from Xiph.org Vorbis 1.4.0 library (as of 2009 and later).

Notes:

  • If you do not want to use the latest AoTuV release you are welcome to wait until the AoTuV changes get merged back into the official Vorbis mainline (This happens for almost every other release of Vorbis).
  • Xiph.org does not maintain the binaries, but rather provides newest updates and releases to libao, libogg, libvorbis, etc so that developers can optimize and build their own compiles

Recommended Vorbis Encoders

(adapted from Recommended Encoder and Settings post compiled by QuantumKnot)

Windows binaries

John33's oggenc2.8 is a special version of the (Ogg) Vorbis encoder. “Features include compression from lossless files (Monkeys Audio, LPAC, FLAC, OptimFROG, WavPack and Shorten – requires presence of decoders), and the ability to specify 'padding' in the headers for subsequent insertion of Tags” (from Ogg Vorbis page at rarewares).


If you prefer a nice drag-and-drop interface, then you can try John33's OggDropXPd (Windows only). “Features include compression from lossless files (Monkey's Audio, LPAC, FLAC and OptimFROG), auto-tagging, renaming of encoded files, setting of advanced encoder parameters, use of VorbisGain tags on decode, Playlist (.m3u) creation, and others” (from Ogg Vorbis page at rarewares).


Note: If the above links do not work, then it is most likely caused by a new version of John33's utilities. In that case, go directly to the Ogg Vorbis page at rarewares.

(OggDropXPd QuickStart guide is here)

Mac OS/X binaries

Users of Mac OS/X can download the following pack of tools:

These tools were compiled by S_O for Mac OS/X.

Linux binaries

AoTuV builds

These builds were compiled with the source code available from aoTuV website mentioned under 3rd party source code. Most of them are static GCC 4 compiles. The the last two versions include libkate, which maybe used for subtitles and some include stdin support with FLAC 1.2.1 if you want to transcode directly from the pipe to Vorbis in Linux.

This static GCC 4 binary was compiled by The_Sven.
This static GCC 4 binary was compiled by QuantumKnot.
This static GCC 4 binary was compiled by artfwo.
This static GCC 4 binary was compiled by artfwo.
This static GCC 4 binary was compiled by artfwo.
This static GCC 4 binary was compiled by artfwo.

Notes:

  • The SSE optimized compile above requires that you have git installed on your Linux distro. Git is a source code revision system. Git can be installed from the source tarballs or via a package manager i.e Synaptic in Ubuntu for instance.

Reference builds

Compiling the latest reference build for Linux requires all dependencies have been met. These dependences include the newest versions of libao, libogg, libvorbis, and vorbis-tools., which can all be downloaded from Xiph.org. Each library should be compiled in the above order in order to satisfy each package to build the latest release of vorbis-tools. libkate is optionally included.

3rd party source code

These are other (Ogg) Vorbis encoders that were tuned by 3rd party developers (outside of Xiph.Org).

The source-code contains modifications mostly to the psychoacoustics model and bitrate allocation, i.e psy.c (aotuv_hf_ weighting, Line 287)
aoTuV Release 1 series releases – The previous recommended encoders
These versions (re-branded from beta 4.51) are an improvement over aoTuV beta 4, which although based on libvorbis 1.1.1, give better quality at low to medium bitrates. Since beta 4, aoTuV includes a -q -2 option for the lowest bitrate. According to forum member guruboolez' latest listening test on classical music, aoTuV beta 4 performed magnificently well at -q 6!! (see Aoyumi's website above for more information)
Many Hydrogenaudio enthusiasts report that Release 1 gives even better quality for low bit-rates. -q 1.5 works for streaming, even good enough.
aoTuV Beta 5 series releases – The recommended encoders
These versions are the latest tuned versions. They further improve low-bitrate encoding, without sacrificing compression. They are now the recommended Vorbis encoders at Hydrogenaudio.
See Compiling aoTuV for information on how to compile it for Linux.

Optimized binaries

These are highly optimized encoders developed by the Ogg Vorbis Acceleration Project codenamed Lancer. They are much faster than the standard binary builds having negligible to nearly no effects on audio quality. These include sped-up routines, i.e mdct.c

These optimized encoders where are one point rapidly changing, as BlackSword found new ways to accelerate, and in the process uncovers new bugs. Please check the “Lancer homepage” link below for the last suite release. They have not been updated since 2006. Developers are better off building their own compiles by downloading the source code from the website above and optmizing them for different architectures. The builds on the website above use old AoTuV builds proceed with caution if you decide to use these.

On Lancer homepage you can find older versions of Oggenc, OggDropXPd, and Dynamic Link Libraries, with optimizations for SSE, SSE2, SSE3 and multi-threading instruction sets.

Note: Output of Lancer may be slightly different from output of 'standard' aoTuV. This is due to the difference of floating-point rounding: Lancer uses 64-bit SSE instructions, in contrast to the standard aoTuV use of 80-bit FP instructions. The output difference between both binaries should not be audible at all. In fact, tests have proven that playback of Lancer's output is indistinguishable from playback of standard aoTuV output.

Forum member The_Sven has produced an aoTuV Beta 5 binary build of Lancer (as of 2009) with optimized SSE instruction sets that has been compiled for Linux. It runs about 3x faster compared to the standard AoTuv Beta 5 build and is a 32-bit x86 binary.

This static GCC 4 binary was compiled by The_Sven.

Released Binaries

How do I know which encoder was used to make this particular (Ogg) Vorbis file?

Using either the ogginfo program or file info in your player, you can tell from the vendor tag:

If you aren't interested in the latest compiles feel free to use Vorbis 1.x.x libraries.

Vendor Tag Encoder Note
Xiphophorus libVorbis I 20000508 1.0 Beta 1 or Beta 2
Xiphophorus libVorbis I 20001031 1.0 Beta 3
Xiphophorus libVorbis I 20010225 1.0 Beta 4
Xiphophorus libVorbis I 20010615 1.0 RC1
Xiphophorus libVorbis I 20010813 1.0 RC2
Xiphophorus libVorbis I 20010816 (gtune 1) 1.0 RC2 GT1
Xiphophorus libVorbis I 20011014 (GTune 2) 1.0 RC2 GT2
Xiphophorus libVorbis I 20011217 1.0 RC3
Xiphophorus libVorbis I 20011231 1.0 RC3
Xiph.Org libVorbis I 20020717 1.0
Xiph.Org/Sjeng.Org libVorbis I 20020717 (GTune 3, beta 1) GT3b1
Xiph.Org libVorbis I 20030308 Post 1.0 CVS
Xiph.Org libVorbis I 20030909 1.0.1
Xiph.Org/Sjeng.Org libVorbis I 20030909 (GTune 3, beta 2) EXPERIMENTAL Experimental GT3b2
Xiph.Org libVorbis I 20031230 (1.0.1) Post 1.0.1 CVS
Xiph.Org/Sjeng.Org libVorbis I 20031230 (GTune 3, beta 2) GT3b2
AO; aoTuV b2 [20040420] (based on Xiph.Org's 1.0.1) aoTuV Beta 2
Xiph.Org libVorbis I 20040629 Xiph.Org Vorbis 1.1 or Xiph.Org Vorbis 1.1 RC1
Xiph.Org libVorbis I 20040920 Xiph.Org Vorbis 1.1 with impulse_trigger_profile
AO; aoTuV b3 [20041120] (based on Xiph.Org's libVorbis) aoTuV Beta 3
Xiph.Org libVorbis I 20050304 Xiph.Org Vorbis 1.1.1 or Xiph.Org Vorbis 1.1.2
AO; aoTuV b4 [20050617] (based on Xiph.Org's libVorbis) aoTuV Beta 4
BS; Lancer [20050709] (based on aoTuV b4 [20050617]) Lancer based on aoTuV Beta 4
AO; aoTuV b4a [20051105] (based on Xiph.Org's libVorbis) aoTuV Beta 4.5
AO; aoTuV b4b [20051117] (based on Xiph.Org's libVorbis) aoTuV Beta 4.51
BS; Lancer [20051121] (based on aoTuV b4b [20051117]) Lancer based on aoTuV Beta 4.51
AO; aoTuV pre-beta5 [20060321] (based on Xiph.Org's libVorbis) aoTuV Beta 5
BS; Lancer(xxx) [yyyymmdd] (based on aoTuV b4b [20051117]) Processor-specific Lancer based on aoTuV Beta 4.51 (1)
BS; Lancer(xxx) [yyyymmdd] (based on aoTuV Release 1 [20051117]) Processor-specific Lancer based on aoTuV Release 1 (1)

(1) Starting 2006-05-06, BlackSword provides accelerated versions for different processors. See the Lancer page for more information.

Recommended Encoder Settings

This is applicable to files encoded in stereo only NOT 5.1/7.1. Refer to the table below. For best results, start at -q 2 and ABX your way up.
Ultimately, the best -q setting will depend on your specific needs. Feel free to experiment.

General Command Line Usage:

oggenc -q n inputfile.wav 
where n is a number from -1 to 10, fractions accepted (using comma or period, depending on where the tool is compiled)
e.g. oggenc -q 5 inputfile.wav

Notes:

  • Most standard oggenc binaries can input lossless FLAC, WavPack, etc files as well (depending upon oggenc version)
  • The current oggenc and libvorbis API do not support the capabilities of "bitrate peeling". Peeling was a "proof-of-concept" idea that would reorder VQ residues by importance allowing the encoder to trim them off thus reducing the bitrate without any loss of quality. Due to the way the codebooks are structured in Vorbis I codec this is not possible and will produce suboptimal files. It will be reconsidered in theoretical working model of Vorbis II called "Ghost" someday.


Switch VBR target
(kbit/s)
VBR range
(kbit/s)
Channel
Coupling
Noise
Normalization
Note
-q -2 ~32 ~32 – ~64 point/lossless yes (1)
-q -1 ~48 ~48 – ~64 point/lossless yes (2)
-q 0 ~64 ~64–~80 point/lossless yes
-q 1 ~80 ~80 – ~96 point/lossless yes (3)
-q 2 ~96 ~96 – ~112 point/lossless yes
-q 3 ~112 ~112 – ~128 point/lossless yes
-q 4 ~128 ~128 – ~160 point/lossless no
-q 5 ~160 ~160 – ~192 point/lossless no (4)
-q 6 ~192 ~192 – ~224 lossless no
-q 7 ~224 ~224 – ~256 lossless no
-q 8 ~256 ~256 – ~320 lossless no
-q 9 ~320 ~320 – ~500 lossless no
-q 10 ~500 ~500 – ~1000 lossless no


(1) Only supported on aoTuVb3 and newer
(2) Bitrate of 48 kbit/s is only for aoTuVb3 and newer. Earlier versions and Xiph.org versions use a bitrate of 45 kbit/s
(3) Reports seem to indicate that aoTuV Release 1 -q 1 provides good quality for streaming
(4) Most users agree -q 5 achieves transparency, if the source is the original or lossless. It is not transparent in the case of transcoding from lossy source (strongly frowned upon)

If you need to know what lowpass settings are used for each quality level, see this HA thread. (It is not recommended that you adjust these)

Advanced Encoder Settings

Reducing pre-echo

This is applicable to all all binaries including the official mainline reference encoder and the AoTuV forks

Generally, using the recommended settings above will give the best quality. There may be cases where Vorbis 1.x.x will fail to reproduce sharp attacks or transients in your music, causing pre echo. In which case, you can use the impulse_noisetune advanced encode switch:

General Command Line Usage:

oggenc -q n --advanced-encode-option impulse_noisetune=p inputfile.wav
where p is a number from 0 to -15
e.g. oggenc -q 5 --advanced-encode-option impulse_noisetune=-5 inputfile.wav

Note that the lower the number (toward -15) for impulse_noisetune, the higher the bitrate will fluctuate in passages of music filled with transient attacks (and the final average bitrate may be much higher than the nominal). Therefore, you should try a small value to start off (say -5) and see if you get acceptable quality. If not, tweak it lower.

Reducing noise due to microattacks

This is only valid for some older Vorbis encoders that are marked as having impulse_trigger_profile. It was considered "experimental" when it was implemented by forum member QuantumKnot in AoTuV Beta 4.51 and can still be used for experimentation purposes.

There are certain parts of some types of music, called “microattacks”, where Vorbis will produce a noise (sort of like a puff of steam), which is due to inaccuracies in the block-switching algorithm (which can't be corrected). Due to the fact the attacks are so fine and close together, Vorbis doesn't switch to impulse short blocks enough, thus “smearing” the reproduction.

By default (with no additional switches), Vorbis selects a different profile for block switching (lower means less switching, higher means more switching) and the default values are shown in the table below:

Quality Profile   Quality Profile
-1 0 5 2.5
0 1 6 2.7
1 1 7 3
2 1.5 8 3.7
3 2 9 4
4 2 10 4

If you encounter this problem on microattacks, you may try the impulse_trigger_profile advanced encode switch, which will change (increase) the profile to your desired value.

oggenc -q n --advanced-encode-option impulse_trigger_profile=r inputfile.wav
where r is a number from 0 to 4, fractions accepted.

Notes:

  • There is the possibility that relaxed block switching (higher profiles) may cause other quality problems and create suboptimal Vorbis files. Please use it sparingly and with caution. If in doubt, leave impulse_trigger_profile on default (that is, don't use it at all)
  • Setting the profile too high will make Vorbis switch to impulse short blocks more often, which will lead to higher bitrate fluctuations. Please be conservative about how you intend to use it.
  • This setting has no effect on reducing the level of pre-echo. It can be said to only reduce the likelihood of pre echo, but the amount of pre-echo is tuned using the impulse_noisetune switch instead.
  • You may try profiles 5 and 6 as substitutes for 3 and 4. Both Profiles 5 and 6, came from 3 and 4 in GT3b2.


You can use both impulse_noisetune and impulse_trigger_profile at the same time, but you will need separate switches, e.g.

oggenc -q n --advanced-encode-option impulse_noisetune=p
  --advanced-encode-option impulse_trigger_profile=r inputfile.wav

Enabling and disabling Vorbis 5.1/7.1 Channel Coupling for Use in Mainline & Google's WebM encoder

This applicable to the latest reference encoder Vorbis 1.4.0 (as of March 2010 and later). It is NOT applicable to the FFMPEG encoder used in WebM which is buggy and should be dropped all together in favor of libvorbis if you are developer or hobbyist compiling WebM from the source in Linux with gcc or Windows with cygwin!

Vorbis generally uses a mixture of very complex channel coupling modes known as "Elliptical" (Point Coupling) and "Dipole" (Phase Coupling). These models are used to reduce the bitrate while taking into account the phase angle and magnitude of waveform respectively. The Point Coupling method is used more frequently to reduce the angle at frequencies where the ear is insensitive to phase.

There maybe times where you want to disable channel coupling so that it just uses a "lossless coupling" model with sacrificing any loss of the surround image. You can optional do this in 5.1 Vorbis in the reference encoder by typing in

oggenc -q n --advanced-encode-option disable_coupling inputfile.wav

Notes:

  • It is generally better to let the encoder use coupling whenever possible. It is NOT recommended you disable it, but if necessary it is possible to do so.